GeDa: Improving training data with large language models for Aspect Sentiment Triplet Extraction

Weixing Mai, Zhengxuan Zhang, Yifan Chen, Kuntao Li, Yun Xue

Abstract: Highlights•Rapid, accurate training data improvement for aspect sentiment triplet extraction.•Two novel types of prompts to guide large language models to synthesize new data.•An iterative strategy for selecting targeted data for diversified model architectures.•Experiments show our framework boosts model performance by 5%＋ with 500 extra data.